Large Language Models: What They Are, Examples, and Open-Source Disadvantages

Latest Update:

Nearly 67% of organizations worldwide use generative AI products powered by LLMs to work with human language and produce content. By 2025, 750 million applications will be using LLMs, with 50% of digital work expected to be automated through apps using these models. Therefore, a proper understanding of large language models can help you secure a job in AI-driven sectors. 

Large language models (LLMs) are advanced AI systems combined with deep learning and algorithms to process and generate human-like texts. Some prominent LLM examples include proprietary, open-source, and specialized LLMs. 

To understand what are large language models, you need to understand transformer architectures and the training of massive datasets for service automation and content generation. In addition, proficiency in fine-tuning models and integration with cloud platforms like AWS can enhance the operational efficiency of LLM models. 

This blog will explore some of the top LLMs and their practical uses that can automate your enterprise operations in 2025.

What are Large Language Models? Origin and Core Concepts

Large language models (LLMs) are artificial intelligence (AI) designed to process, understand, and generate human language. They enable them to perform tasks like language translation, content generation, and answering complex questions. LLMs are built using neural networks, specifically transformer models, deep learning architectures capable of processing large-scale language data. 

If you want to learn essential skills to help you build scalable LLM models, the following courses can help you succeed. 

• Executive Diploma in Machine Learning and AI with IIIT-B
• Generative AI Foundations Certificate Program
• Advanced Generative AI Certification Course

Here’s a comparison of LLMs to traditional natural language processing (NLP) models:

• Scale: Large language models like GPT-3 or GPT-4 are trained on billions of parameters, enabling them to process much more information than earlier models. Traditional NLP models, by contrast, often handled simpler tasks with far fewer parameters.
• Context handling: LLMs excel at understanding and maintaining context over long text spans, whereas traditional NLP models struggled with maintaining coherence across significant texts, often losing track of earlier parts of the conversation or document.
• Output quality: LLMs generate more coherent, contextually aware, and diverse output, enabling them to create human-like text. Traditional NLP models often produce rigid and formulaic responses.

Here are some key milestones in the development of large language models that you must know:

• BERT (Bidirectional Encoder Representations from Transformers): Released by Google in 2018, BERT revolutionized NLP by introducing a method to pre-train language models in both directions, improving understanding of context. 
• GPT-1 (Generative Pretrained Transformer 1): Introduced by OpenAI in 2018, ChatGPT was one of the first models to demonstrate the power of large-scale unsupervised pre-training. The model learned language patterns from vast amounts of text data without specific task supervision. 

The early milestones laid the foundation for sophisticated models such as GPT-3 and GPT-4, which are now capable of accurately performing complex tasks. 

Now, let’s explore how large language models work comprehensively. 

How Do LLMs Work? Training, Data, and Parameters

Large language models use advanced machine learning (ML) techniques based on transformer architectures. These models are trained on massive datasets typically sourced from diverse text-rich data such as books, articles, and web content. To understand what are large language models comprehensively, you must assess the training process, which involves self-supervised learning, which allows LLMs to generate human-like texts. 

Here are some of the key components for LLM training:

1. Self-Supervised Learning

Self-supervised learning is key for LLM training, where the model learns to predict parts of data based on other parts, without labeled data. In the case of LLMs, the model predicts the next token in a sequence or fills in masked tokens based on the surrounding context.

• Masked Language Modeling: This technique, commonly used in BERT, involves masking tokens in the text, and the models predict these missing tokens. 
• Autoregressive modeling: GPT-3 is a good example, generating text sequentially by predicting the next word based on the preceding context. 

Use case:

Self-supervised learning enables LLMs like GPT-3 to produce high-quality articles, blog posts, or creative writing in content generation. The model predicts the next word and phrase based on the context of your provided sentence, ensuring coherence and relevancy in generated text. 

2. Datasets

Training an LLM requires large, diverse datasets to ensure the model can generalize well across different contexts. These datasets are crucial in helping the model capture language patterns, syntactic structures, and semantic relationships.

• C4 (Colossal Clean Crawled Corpus) is a large-scale, web-scraped dataset used for pre-training. It provides diverse text data across multiple domains.
• BooksCorpus: It is a dataset containing over 11,000 books offering formal and structured language data. 
• Scientific papers and Wikipedia: Domain-specific datasets like Wikipedia help understand specialized languages necessary for technical content generation and answer your domain-specific queries. 

Use case:

You can train LLMs on domain-specific texts such as legal texts. A fine-tuned LLM can help you review a contract and identify payment terms, confidentiality, and dispute resolution clauses. Moreover, the model can suggest edits for clarity or compliance with legal standards. 

3. Parameters

The number of parameters in an LLM determines its capacity to capture complex patterns and relationships within language. These parameters are the internal weights that define how the model processes the input data you provide and generates output.

• GPT-3 contains 175 billion parameters, generating highly coherent, contextually accurate responses across various domains.
• GPT-4 likely surpasses GPT-3 in parameter count, further enhancing its ability to understand and generate complex language.

Use case:

Large language models with large parameter counts can provide personalized content recommendations, such as tailored news articles and product recommendations. For example, if you work in an online news platform, you can use LLMs to analyze users’ past reading habits and recommend articles. 

4. Data-Parameter Interaction

There is a direct relationship between the amount of training data and the number of parameters in an LLM. A model with billions of parameters requires massive datasets to avoid overfitting and to ensure the model can generalize effectively across diverse tasks.

• Data volume: LLMs like GPT-3 are trained on datasets with hundreds of billions of tokens, allowing the model to learn diverse language patterns.
• Data diversity: A mix of general and specialized datasets is required to ensure the model can effectively handle various language tasks. 
• Parameter Scale: The larger the parameters, the more data the model requires to learn effectively without overfitting.

Use case:

An LLM can summarize a lengthy quarterly financial report into a brief overview if you are in a corporate setting. The interaction between large datasets and parameters allows the model to retain key details while reducing redundancy. Your company can see an increase in revenue of 15%. 

5. Role of Computing Power

Training LLMs demands massive compute power, usually provided by GPUs and TPUs. These accelerators enable parallel processing, where computations are distributed across multiple hardware units.

• Data parallelism: In training, datasets are divided into smaller batches and processed across multiple GPUs. It enables you to train your model efficiently with parallel processing capabilities. 
• Model parallelism: Large language models such as GPT-3 have billions of parameters, making fitting them in a single GPU difficult. Model parallelism splits the model across multiple GPUs, each handling a different part of the model, facilitating efficient large-scale data handling.

Use case:

In autonomous vehicle systems, LLMs utilize powerful computational resources to interpret sensor data, understand road signs, and generate driving decisions. An LLM on your vehicle’s onboard system can interpret voice commands, integrating language understanding and real-time spatial awareness. 

6. Fine-Tuning

After pre-training on large datasets, LLMs undergo fine-tuning to make them specialized in domain-specific activities. Fine-tuning involves training the model on a smaller, domain-specific dataset, allowing it to adapt its learned knowledge to particular tasks. 

Such tasks include:

• Customer support: You can fine-tune an LLM on customer interaction data to provide accurate responses in a customer service context. 
• Healthcare: You can also fine-tune medical texts, which helps the model generate relevant and accurate responses for healthcare-related queries. 

Use case:

You work for a financial institution in Delhi, where you use LLMs to evaluate the risks of loan applicants by analyzing financial historical data. Fine-tuning LLMs on financial data allows you to accurately assist in credit scoring, fraud detection, and investment recommendations. 

7. Transformer Architecture

Transformers utilize a self-attention mechanism that allows the model to weigh the importance of different words in a sentence relative to each other. It is crucial to understand long-range dependencies in the text you provided.

• Self-attention: This mechanism enables the model to simultaneously process all words in a sentence and learn which words are most relevant to others, even over long distances.
• Parallelization: Unlike older architectures like RNNs and LSTMs, transformers process all tokens in parallel, which speeds up training and enables more efficient learning.
• Scalability: The transformer’s structure is inherently scalable, allowing for the training of massive models like GPT-3 without a significant loss in performance.

Use case:

You are working with a news agency where a transformer architecture can automatically classify articles depending on politics or entertainment. The transformer architecture allows your LLM to understand complex text classifications where categorization depends on predefined categories. 

Now, let’s look at some disadvantages of open-source large language models. 

Disadvantages of Open Source Language Models: Key Issues

Open-source language models such as GPT-Neo, BERT, and other pre-trained architectures have become popular due to their accessibility and customization flexibility. However, these models come with notable disadvantages of open-source large language models, such as misuse risks, lack of content filtering, and intellectual property issues.

Here are some of the core disadvantages of open source language models:

1. Risks of Misuse and Security Threats

Open-source models are publicly available and can be freely accessed, making them vulnerable to exploitation by malicious actors. These models can generate harmful or unethical content without proper controls, such as deepfakes, misinformation, or social engineering attacks.

Effect: With models openly available on platforms like GitHub, there is no centralized oversight, making it challenging to safeguard model integrity. You must deploy additional security measures, often requiring complex integrations with platforms like Docker and cloud environments like AWS to detect misuse. 

Example: GPT-2, when initially released, was a perfect example of how the disadvantage of open-source large language models​ can lead to misuse. After its release, it was used to generate fake news articles, making it evident that LLMs could easily be exploited for malicious purposes without safeguards.

2. Lack of Content Filtering and Safeguards

Open-source language models generally lack built-in content filtering systems, which means they can produce biased, offensive, or harmful outputs. These models may generate inappropriate or unethical content without additional safeguards, leading to reputational damage or legal challenges.

Effect:

You must implement complex content filtering systems with tools like TensorFlow or integrate custom scripts in Python to process outputs. This added burden does not exist when using proprietary models with built-in safeguards.

Example: 

When you use BERT for tasks like sentiment analysis or content generation, biases tend to be generated. A fine-tuned BERT model for an HR application could unintentionally reinforce gender biases, a disadvantage of open-source large language models​ when deployed in sensitive fields.

3. Intellectual Property and Copyright Problems

Open-source models are trained on vast datasets scraped from publicly available sources, which may include copyrighted content. As a result, your generated outputs could violate intellectual property rights by mimicking or reproducing protected content.

Effect: Enterprises must implement mechanisms for detecting and preventing unintentional copyright infringement with extra care. You can integrate third-party plagiarism detection software or build custom solutions in Java or Python.

Example: You are working in an enterprise using GPT-Neo for content generation in marketing campaigns. You may find content closely resembling copyrighted texts, which is a clear disadvantage of open-source large language models​. 

4. Lack of Professional Support and Documentation

Open-source models generally lack official customer support or comprehensive documentation. Developers rely on community forums, GitHub discussions, or external consultants, which may not always provide the timely and accurate support for mission-critical applications.

Effect: In this context, the disadvantage of open-source large language models​ is that the lack of structured support leads to longer troubleshooting times. While deploying these models, you must implement this with internal teams proficient in machine learning frameworks like TensorFlow or PyTorch.

Example: You are using GPT-2 for a chatbot integration project where you encounter issues in fine-tuning and scaling. Without professional support from the model's developers, you would have to rely on documentation or community contributions, which may not provide sufficient troubleshooting assistance.

5. Computational Cost and Resource Requirements

Large open-source models, like GPT-Neo and GPT-2, require significant computational resources to train and deploy. Running these models at scale often necessitates high-performance hardware such as GPUs or TPUs, along with specialized environments for deployment, such as Kubernetes. 

Effect: Deploying open-source models on platforms like AWS or managing resources via Docker for containerization incurs significant costs. Small businesses may struggle with these expenses, a critical barrier to large-scale deployment.

Example: Fine-tuning GPT-3 on a specific dataset requires powerful GPU clusters or cloud services, which can become expensive for your company. The disadvantage of open-source large language models​ is that while the model's code is free, the resources required to deploy them efficiently are not.

Also read: 17 AI Challenges in 2025: How to Overcome Artificial Intelligence Concerns?

Let’s explore some benefits and typical uses of large language models. 

Benefits and Common Uses of Large Language Models

Large Language Models (LLMs) like GPT-3 and T5 have become fundamental tools in NLP, due to their utilization of transformer architectures. To understand what large language models are and their benefits, you must explore complex relationships within texts and prompts. 

Here are some benefits and common uses of large language models:

Benefits 

• Versatility across different applications: LLMs are trained on diverse datasets using unsupervised and self-supervised learning techniques, which makes them adaptable to various industries.
  • Example: You are working in the healthcare sector in Hyderabad, where LLMs can analyze clinical data and identify potential drug interactions. You can also fine-tune medical data and use spaCy to recognize named entities. 
• Improvements in contextual understanding: LLMs capture context through self-attention layers, which enable them to process long-range dependencies and relationships across text sequences. By utilizing large-scale pre-trained embeddings, they perform better in maintaining context during multi-turn interactions or long-text summarization.
  • Example: You work in a customer service sector, where you can use LLMs to understand previous queries and stateful architectures to track history during conversations. 
• Enabling faster innovation across fields: LLMs enable rapid prototyping in fields like AI research, where they assist in hypothesis generation and data-driven insights. Using transfer learning, fine-tuning on domain-specific datasets accelerates innovation in industry applications and academic research.
  • Example: You work in financial institutions, using LLMs to automate risk analysis and process large volumes of financial documents. You also use Optuna to optimize hyperparameters and generate automated summaries for regulatory compliance. 

Common uses

• Text and content generation: LLMs like GPT-3 are trained using maximum likelihood estimation (MLE) and causal language modeling to generate high-quality, coherent text based on a given prompt. Fine-tuning these models on specific domains allows for generating context-aware content.
  • Example: LLMs are fine-tuned on specific keywords and market trends for SEO-optimized content to generate relevant and engaging posts. You can also utilize TF-IDF and latent semantic indexing (LSI) for keyword optimization.
• Programming assistance and code completion: LLMs can be utilized in Integrated Development Environments (IDEs) to complete code using tokenization and semantic similarity algorithms. These models are fine-tuned on GitHub repositories and integrate with IDEs via APIs to provide contextually accurate code snippets.
  • Example: GitHub Copilot, powered by Codex, can generate entire functions or suggest code completions based on your initial input. You can also use transformers to predict and generate Python, JavaScript, and Ruby code suggestions.
• Translation and language services: LLMs can perform high-quality translation using multi-task learning and training on large parallel datasets. They utilize encoder-decoder architectures to map source text to target languages while retaining context and meaning.
  • Example: Google Translate uses transformer-based models to provide contextually relevant translations, utilizing back-translation and pre-training on millions of multilingual datasets to handle idiomatic expressions.

Also read: 5 Significant Benefits of Artificial Intelligence [Deep Analysis]

Now, let’s explore what large language models​ are to various examples:

Large Language Model Examples: Leading LLMs in 2025

LLM models are classified into proprietary, Open-Source, and specialized LLMs. Each type offers distinct capabilities, such as conversational agents, code generation, or medical text analysis. 

Here is a comprehensive overview of some of the large language models examples present in 2025:

Proprietary LLMs

Proprietary LLMs are developed and controlled by enterprises. You can access these models through APIs or subscription services for reliable performance.

Here are some proprietary large language models examples​: 

OpenAI's GPT Series (GPT-3, GPT-4, ChatGPT)

Source: openai.com

OpenAI’s GPT-3 and GPT-4 are among the most advanced proprietary LLMs, capable of producing human-like text generation, summarization, and question answering. With 175 billion parameters in GPT-4, these models are prominent large language models examples

• GPT-4 supports text and image input, making it versatile for applications beyond pure text, which makes it a prominent large language models examples in enterprises. 
• GPT-4 is optimized for performance and scale due to the presence of 175 billion parameters, enabling content generation in a human-like manner. 
• API access through OpenAI’s APi allows you to integrate the model into third-party applications for streamlined product deployment. 
• Understanding of natural languages makes it unique for text generation and summarization. 

Use case:

In customer service, GPT-4 can enhance chatbot performance by accurately understanding customer inquiries and providing personalized responses based on historical data. By integrating GPT-4 via OpenAI's API, you can automate customer support, reducing operational costs while increasing customer satisfaction through real-time, intelligent assistance.

Google’s Gemini 

Source: gemini.google.com

Gemini is part of Google’s suite of multimodal LLMs, designed to work across text and image inputs. It is highly versatile and optimized for handling tasks that require understanding and generating content in multiple formats.

• The LLM processes text, images, and video, enabling a more comprehensive understanding of content.
• Available for enterprise applications through Google Cloud and uses pathways to scale across various tasks. 
• It utilizes the pathways system, enabling Google’s LLMs to work more efficiently across multiple tasks and improving model performance and scalability.

Use Case:

In digital marketing, Gemini can generate integrated campaigns that combine text and visuals. Using Gemini’s ability to process text and image inputs, you can automate ad copy creation and corresponding visuals, significantly reducing time spent on content production while ensuring cohesive and targeted messaging.

Google’s PaLM

Source: ai.google

PaLM (Pathways Language Model) is a general-purpose LLM from Google designed to scale across multiple NLP tasks. It uses the Pathways architecture to allow better efficiency and model performance.

• PaLM can handle complex NLP tasks such as reasoning and text generation.
• It enables the training of one model to handle multiple tasks simultaneously, optimizing performance. It is available through Google Cloud’s AI platform. 
• The LLM is used for complex NLP tasks like automated reasoning, content generation, and multilingual document analysis.

Use case:

Legal firms use PaLM for multilingual legal document analysis. With PaLM's capabilities, you can automate the translation of contracts and the study of legal texts across languages. In addition, it streamlines international contract review and reduces the need for manual translation services.

Anthropic’s Claude

Source: anthropic claude

Claude is a series of LLMs from Anthropic designed to focus on safety, ethics, and transparency. These models are built to understand and generate human-like text while avoiding harmful or biased outputs, making a prominent large language models examples in enterprises. 

• Claude allows you to control how the model responds, ensuring it aligns with ethical guidelines and organizational standards.
• The models have mechanisms to mitigate harmful responses and avoid ethical issues.
• You can use these models that require responsible, ethical language generation, such as chatbots and customer services. 

Use case:

Mental health platforms use Claude to provide safe, compassionate, and ethical support through conversational agents. By integrating Claude, you can ensure that AI responses are empathetic and secure for users while adhering to regulatory guidelines and avoiding harmful or biased language.

Also read: 28+ Top Generative AI Tools in 2025: Key Benefits and Uses

Open-Source LLMs

Open-source LLMs are publicly available, allowing you to modify and fine-tune the models for specific applications. These models are popular in academic research and industries looking to customize solutions without being locked into proprietary ecosystems.

Here are some open-source large language models examples​: 

Meta's LLaMA 2, 3

Source: ai.meta.com

LLaMA (Large Language Model Meta AI) is Meta's open-source LLM, designed for efficiency in NLP tasks. It is optimized for research and production environments, offering competitive performance and lower resource consumption.

• LLaMA models are designed to be more computationally efficient while maintaining high performance for your enterprise applications.
• Despite being smaller, LLaMA provides strong results for all your parameters on standard NLP benchmarks.
• LLaMa is available on GitHub, allowing you to customize the model for specific applications.

Use case: 

Research institutions use LLaMA 3 to conduct large-scale NLP experiments in computational linguistics. The platform lets you rapidly train models with minimal hardware investment, especially for specialized research projects like semantic analysis.

Mistral 7B

Source: mistral.ai

Mistral is an open-source model designed for efficiency and high performance. Mistral 7B is one of its most notable versions. It is optimized for tasks requiring low-latency inference and can handle complex language tasks despite having fewer parameters.

• Mistral is designed for operations that demand faster processing and fewer computational resources.
• The LLM is available under an open-source license, making it accessible for research and production use.
• Mistral is designed for applications requiring rapid text generation, making it ideal for custom interactions and systems with limited resources.

Use case:

Mistral 7B is used in e-commerce platforms for real-time customer support chatbots. It is critical if you are working with low-latency processing to deliver instant answers and maintain high customer satisfaction during peak times.

Mixtral

Source: mistral.ai

Mixtral is an open-source LLM known for being efficient and lightweight while offering high performance in generating text and handling NLP tasks. It is optimized for both resource efficiency and speed.

• Mixtral is designed to be faster and require fewer resources than larger models like GPT-3.
• The LLM offers multilingual support and is openly available for research, customization, and production. 
• Ideal for applications that need real-time or low-latency responses, like chatbots or interactive assistants.

Use case:

Online retailers deploy Mixtral to handle real-time customer inquiries, where high traffic and immediate responses are crucial. The low-latency feature ensures you for smooth interaction even during peak shopping periods.

Falcon

Source: falconai.com

Falcon is an advanced open-source LLM designed for both efficiency and performance. It is suitable for a wide variety of NLP tasks, particularly those that require real-time response generation.

• Falcon is designed to work with multiple languages, providing multilingual support for cross-lingual tasks.
• The LLM is available for modification and integration with custom solutions. 

Use case:

Falcon powers multilingual customer support systems for global enterprises. It allows you to efficiently interact with customers in multiple languages without needing separate teams for each region.

BLOOM

Source: bloomai.co

BLOOM is an open-source model designed for high-quality, long-form text generation. It is built to support tasks requiring coherent and detailed content, such as articles, papers, and creative writing.

• BLOOM is beneficial in creating structured and coherent text for long-form applications.
• The LLM is trained on diverse texts to ensure adaptability across various domains and is available as an open-source LLM for customization and development. 

Use case:

BLOOM is widely used in content marketing by agencies and enterprises to automate the creation of detailed blog posts. It also allows you to prepare research summaries and product descriptions, significantly reducing content production time.

Specialized LLMs

Specialized LLMs are tailored for tasks such as code generation or medical applications. These models use multilingual programming with JavaScript and Python for accurate outputs. 

Here are some specialized large language model examples​: 

Codex

Source: generaltranslation.com

Codex is a specialized LLM focused on generating and completing code. These models have been trained on large codebases, making them prominent large language models examples for software development and debugging. 

• Codex can generate code based on natural language descriptions, making it ideal for your product development scenario. 
• The LLM supports various programming languages, including Python, JavaScript, and Ruby.
• The platform helps you by providing real-time code suggestions, completing functions, and debugging code.

Use case: 

GitHub Copilot, powered by Codex, significantly speeds up the development process by providing real-time code suggestions. Copilot automatically suggests functions as you write code, completes code snippets, and helps debug issues, making software development more efficient and less prone to error.

CodeGen

Source: codegen.com

CodeGen is a specialized LLM for programming and code completion. Trained on a large corpus of programming languages and repositories, it generates syntactically correct code from natural language prompts.

• It provides you with context-aware code generation and recommendations.
• It supports languages such as Python, Java, C++, JavaScript, and more, making it adaptable for various development environments.
• You can integrate it into popular IDEs like Microsoft Visual Studio and JetBrains, enabling error detection and improving workflow efficiency.

Use case: 

CodeGen is integrated into IDEs like Visual Studio, which assists you by suggesting code snippets, completing functions, and providing real-time error feedback. You can also use it to prototype software faster and to troubleshoot, ensuring an efficient and accurate development process. 

MedPaLM

Source: medpalm.com

MedPaLM is an LLM designed for medical applications. It provides insights into clinical texts and assists with decision-making based on medical knowledge.

• These LLMs are fine-tuned on clinical data to handle medical terminology and provide insights into patient care.
• The LLM supports doctors by generating recommendations and summarizing patient data for better clinical decisions. 

Use case:

You are a medical practitioner in a prominent hospital in Mumbai. MedPaLM can automate the analysis process and provide insights into diagnosis and treatment plans. The system analyzes data from various sources, such as medical records, lab results, and clinical notes, to provide actionable recommendations tailored to individual patient needs.

Also read: Top 20 Types of AI in 2025 Explained

Become a Large Language Model Expert Today with AI Expertise!

Large language models such as GPT-4, BERT, and PaLM are revolutionizing industries with advanced text generation, natural language understanding, and process automation. These models use transformer architectures and are trained on vast datasets for content generation, multilingual translation, and more. 

Proprietary LLMs like Google’s Gemini offer reliable performance, but open-source LLMs like Mistral 7B present challenges in data privacy, misuse risks, and resource requirements.

If you want to stay ahead of your peers with industry-relevant LLMs, look at upGrad’s courses that allow you to be future-ready. These are some of the additional courses that can help expand your AI journey.

• Master the Future of Data with Microsoft 365 Copilot
• Generative AI Mastery Certificate for Data Analysis
• ChatGPT for Developers

Curious which courses can help you gain expertise in LLMs in 2025? Contact upGrad for personalized counseling and valuable insights. For more details, you can also visit your nearest upGrad offline center. 

Expand your expertise with the best resources available. Browse the programs below to find your ideal fit in Best Machine Learning and AI Courses Online.

Best Machine Learning and AI Courses Online

Discover in-demand Machine Learning skills to expand your expertise. Explore the programs below to find the perfect fit for your goals.

In-demand Machine Learning Skills

Discover popular AI and ML blogs and free courses to deepen your expertise. Explore the programs below to find your perfect fit.

Popular AI and ML Blogs & Free Courses

Reference Link:
https://keywordseverywhere.com/blog/llm-usage-stats/